Iñigo Perona

An extensive comparative study of cluster validity indices

The validation of the results obtained by clustering algorithms is a fundamental part of the clustering process. The most used approaches for cluster validation are based on internal cluster validity indices. Although many indices have been proposed, there is no recent extensive comparative study of their performance. In this paper we show the results of an experimental work that compares 30 cluster validity indices in many different environments with different characteristics. These results can serve as a guideline for selecting the most suitable index for each possible application and provide a deep insight into the performance differences between the currently available indices.

SEP/COP: An efficient method to find the best partition in hierarchical clustering based on a new cluster validity index

Hierarchical clustering algorithms provide a set of nested partitions called a cluster hierarchy. Since the hierarchy is usually too complex it is reduced to a single partition by using cluster validity indices. We show that the classical method is often not useful and we propose SEP, a new method that efficiently searches in an extended partition set. Furthermore, we propose a new cluster validity index, COP, since many of the commonly used indices cannot be used with SEP. Experiments performed with 80 synthetic and 7 real datasets confirm that SEP/COP is superior to the method currently used and furthermore, it is less sensitive to noise.

Web usage and content mining to extract knowledge for modelling the users of the Bidasoa Turismo website and to adapt it

Abstract The tourism industry has experienced a shift from offline to online travellers and this has made the use of intelligent systems in the tourism sector crucial. These information systems should provide tourism consumers and service providers with the most relevant information, more decision support, greater mobility and the most enjoyable travel experiences. As a consequence, Destination Marketing Organizations (DMOs) not only have to respond by adopting new technologies, but also by interpreting and using the knowledge created by the use of these techniques. This work presents the design of a general and non-invasive web mining system, built using the minimum information stored in a web server (the content of the website and the information from the log files stored in Common Log Format (CLF)) and its application to the Bidasoa Turismo (BTw) website. The proposed system combines web usage and content mining techniques with the three following main objectives: generating user navigation profiles to be used for link prediction; enriching the profiles with semantic information to diversify them, which provides the DMO with a tool to introduce links that will match the users taste; and moreover, obtaining global and language-dependent user interest profiles, which provides the DMO staff with important information for future web designs, and allows them to design future marketing campaigns for specific targets. The system performed successfully, obtaining profiles which fit in more than 60% of cases with the real user navigation sequences and in more than 90% of cases with the user interests. Moreover the automatically extracted semantic structure of the website and the interest profiles were validated by the BTw DMO staff, who found the knowledge provided to be very useful for the future.

SIHC: A STABLE INCREMENTAL HIERARCHICAL CLUSTERING ALGORITHM

SAHN is a widely used agglomerative hierarchical clustering method. Nevertheless it is not an incremental algorithm and therefore it is not suitable for many real application areas where all data is not available at the beginning of the process. Some authors proposed incremental variants of SAHN. Their goal was to obtain the same results in incremental environments. This approach is not practical since frequently must rebuild the hierarchy, or a big part of it, and often leads to completely different structures. We propose a novel algorithm, called SIHC, that updates SAHN hierarchies with minor changes in the previous structures. This property makes it suitable for real environments. Results on 11 synthetic and 6 real datasets show that SIHC builds high quality clustering hierarchies. This quality level is similar and sometimes better than SAHN’s. Moreover, the computational complexity of SIHC is lower than SAHN’s.

Consolidated trees versus bagging when explanation is required

In some real-world problems solved by machine learning it is compulsory for the solution provided to be comprehensible so that the correct decision can be made. It is in this context that this paper compares bagging (one of the most widely used multiple classifier systems) with the consolidated trees construction (CTC) algorithm, when the learning problem to be solved requires the classification made to be provided with an explanation. Bearing in mind the comprehensibility shortcomings of bagging, the Domingos’ proposal, called combining multiple models, has been used to address this problem. The two algorithms have been compared from three main points of view: accuracy, quality of the explanation the classification is provided with, and computational cost. The results obtained show that it is beneficial to use CTC in situations where an explanation is required, because: CTC has a greater discriminating capacity than the explanation extraction algorithm added to bagging; the explanation provided is of a greater quality; it is simpler and more reliable; and CTC is computationally more efficient.

Web mining for navigation problem detection and diagnosis in Discapnet: A website aimed at disabled people

The dramatic increase in the amount of information stored on the web makes it more important to familiarize people with disabilities and elderly people with digital devices and applications and to adapt websites to enable their use by these users. Discapnet is a website mainly aimed at visually disabled people, and navigation is a challenging task for its users. In this context, system evaluation and problem detection become crucial aspects for enhancing user experience and may contribute greatly to diminishing the existing technological gap. This study proposes a system based on web‐mining techniques that collects in‐use information while the user is accessing the web (thus, being a noninvasive system). The proposed system models users in the wild and discovers navigation problems appearing in Discapnet and can also be used for problem detection when new users are navigating the site. The system was tested and its efficiency demonstrated in an experiment involving navigation under supervision, in which 82.6% of a set of disabled people were automatically labeled as having problems with the website.

SAHN with SEP/COP and SPADE, to build a general web navigation adaptation system using server log information

During the last decades, the information on the web has increased drastically but larger quantities of data do not provide added value for web visitors; there is a need of easier access to the required information and adaptation to their preferences or needs. The use of machine learning techniques to build user models allows to take into account their real preferences. We present in this work the design of a complete system, based on the collaborative filtering approach, to identify interesting links for the users while they are navigating and to make the access to those links easier. Starting from web navigation logs and adding a generalization procedure to the preprocessing step, we use agglomerative hierarchical clustering (SAHN) combined with SEP/COP, a novel methodology to obtain the best partition from a hierarchy, to group users with similar navigation behavior or interests. We then use SPADE as sequential pattern discovery technique to obtain the most probable transactions for the users belonging to each group and then be able to adapt the navigation of future users according to those profiles. The experiments show that the designed system performs efficiently in a web-accesible database and is even able to tackle the cold start or 0-day problem.

Inferring Visual Behaviour from User Interaction Data on a Medical Dashboard

(its size and complexity) and its context of use. This results in user interfaces with a high-density of data that do not support optimal decision-making by clinicians. Anecdotal evidence indicates that clinicians demand the right amount of information to carry out their tasks. This suggests that adaptive user interfaces could be employed in order to cater for the information needs of the users and tackle information overload. Yet, since these information needs may vary, it is necessary first to identify and prioritise them, before implementing adaptations to the user interface. As gaze has long been known to be an indicator of interest, eye tracking allows us to unobtrusively observe where the users are looking, but it is not practical to use in a deployed system. Here, we address the question of whether we can infer visual behaviour on a medication safety dashboard through user interaction data. Our findings suggest that, there is indeed a relationship between the use of the mouse (in terms of clickstreams and mouse hovers) and visual behaviour in terms of cognitive load. We discuss the implications of this finding for the design of adaptive medical dashboards.

Modeling the Navigation on Enrolment Web Information Area of a University Using Machine Learning Techniques

This work analyses the navigation in the enrolment web information area of the University of the Basque Country. A complete data mining process shows that successful and failure navigation behaviors can be modeled using machine learning techniques. Unsupervised learning algorithms have been applied on two different domains: URLs visited by the users in each session (navigation sequence) and some interaction parameters extracted from the recorded click-stream (navigation style). Both domains have been used satisfactorily to model the behavior of success and failure navigation sessions achieving more than 78% of accuracy predicting success or failure sessions. Furthermore, the clustering based on the navigation style was able to identify the main characteristics of each type of session and to build a subsystem that enables to detect failure type sessions with high precision.

Global versus modular link prediction approach for discapnet: Website focused to visually impaired people

Web personalization becomes essential in industries and specially for the case of users with special needs such as visually impaired people. Adaptation may very much speed up the navigation of visually impaired people and contribute to diminish the existing technological gap. This work is the first stage of a web mining process carried out in discapnet: a website created to promote the social and work integration of people with disabilities where slow navigation has been detected. Based on observation in-use where behaviours emerge applying a web mining process to server log data, we designed a system to generate user navigation profiles and adapt to the web site through link prediction. Two approaches for user profiling were implemented: a global system built based on the complete database and a modular approach carried out discovering the navigation profiles within different zones. Although both approaches are effective, the modular approach outperforms. When 25% of the navigation of the new user has happened the designed system is able to propose a set of links where nearly 60% of them (2 out of 3) is among the ones the new user will be using in the future. This will definitely make the navigation easier saving a lot of time.